Is Anthropomorphic Design a Viable Way of Enhancing Interface Usability? |
||||||
Chapter One: Introduction |
||||||
Since computing began, many developers have tried to create systems which exhibit human emotions or personality traits. Some authors, such as Shneiderman (Shneiderman & Plaisant, 2004) believe that ascribing human characteristics to an interface is wrong as it may confuse or mislead users, it blurs the boundaries between human and computer and that while some people like it, others find it distracting or unsettling. Despite this, humans do seem to have a primitive urge to anthropomorphise certain objects, such as cars or computers. Extensive research, like the Computers Are Social Actors (CASA) paradigm (Nass, Steuer and Tauber, 1994) has repeatedly proven that users instinctively treat their computers as if they were real people. So, if anthropomorphism is inevitable, should we embrace it? If it is implemented in the right way, will it enhance the user interface experience? This report considers the existing arguments for and against the use of anthropomorphism in interface design and examines why previous attempts at items such as interface agents have generally been unsuccessful. It analyses data gathered from computer users, via a web-based questionnaire, to ascertain to what extent they might view their own computer as a ‘person’ – eg through physical contact, talking to it or expecting things from it – and how they view existing anthropomorphic interfaces. Chapter Four describes how this information was applied to create two versions of a word processing application. One version used an anthropomorphic interface, one did not, but they had the same basic functionality. Both versions were then tested on users in an empirical study to find out which interface style they preferred. The speed and effectiveness of the users actions were also documented, via inbuilt recording methods, in order to assess the usability and productivity-enhancing capabilities of each system. The results of the user tests were then compared and evaluated to find out if anthropomorphism has a viable role in enhancing either usability or the user interface experience or whether it should be avoided at all costs. |
||||||
1.1 Anthropomorphism |
||||||
Anthropomorphism is the attribution of human characteristics or behaviour to non-human things such as animals, natural phenomena or inanimate objects. An ancient and compelling natural trait, it seems that it is impossible for humans not to anthropomorphise. Historically, human characteristics have often been applied to technology. A good example of this is the way in which people anthropomorphise their cars. Plant (1993, cited by Barrow, 2000) has suggested that this may stem from man’s need to assert dominance and show who is in charge. In his paper Moving Relationships: Befriending the Automobile to Remove Anxiety (1999), Jameson Wetmore suggests that anthropomorphism is used to help us assimilate technology into our lives. When attributing a car with human (and often quirky) characteristics an individual is better able to counteract any feelings of fear, threat or confusion: “By giving a person a framework within which to view the situation, it can help a confused or frustrated owner regain a sense of control.” For example: if a driver notices that their fuel gauge says empty when they are driving down the motorway, they may be able to negate the stress of the situation by treating the car as a companion. Wetmore suggests that historically, our approach to technology has usually involved different groups in society negotiating the meaning of a technology until a dominant meaning emerges. The technology then has to be accepted at an individual level in order for it to succeed. Anthropologist Stewart Guthrie (1993: 5) argues that anthropomorphism is also a way for people to understand something which is impossible to explain: “We animate and anthropomorphise because, when we see something as alive or humanlike, we can take precautions. If we see it as alive we can, for example, stalk it or flee. If we see it as humanlike, we can try to establish a social relationship. If it turns out not to be alive or humanlike, we usually lose little by having thought it was.” Guthrie states that our world is uncertain, ambiguous and in need of interpretation. He asserts that to interpret things by disclosing the presence of whatever is most important to us (ie other humans) is a ‘good bet’. He likens his theory to Pascal’s wager, concerning the existence of God: The wager suggests that we might as well believe because if we are eventually proved right, we will gain eternal reward, whereas if it turns out that He doesn’t exist, we would have lost very little. What have we got to lose? This study investigates how Guthrie’s theory may be applied to the issues surrounding the use of anthropomorphism in interface design. – If it is human nature to treat our computer as a social actor and we are able to modify the technology to compliment this behaviour, what indeed, have we got to lose? Ben Shneiderman (personal communication, 2005) claims that creating a humanlike interface is counterproductive and that human-human interaction is rarely a good source of inspiration for human-computer interaction. On the other hand, champions of the anthropomorphic approach, such as Clifford Nass are continuing to investigate the CASA (Computers Are Social Actors) (Nass Steuer and Tauber, 1994) paradigm and the application of humanlike interfaces in their current research. |
||||||
1.2 Anthropomorphism and Computing |
||||||
In 1950 the mathematician Alan Turing wrote the paper ‘Computing Machinery and Intelligence’ and developed the Turing test - a proposal for a method of testing a machine's capability to perform human-like conversation. To this day, an annual competition based on the Turing test, called the Loebner Prize, is awarded to the ‘chatbot’ computer program which is considered to be the most likely to fool the judges into thinking it is human. At around the same time computing pioneer John von Neumann was one of the first people to liken a computer to the human mind. The human tendency to believe that the human brain is like a computer, and therefore, that a computer can have human qualities, is known as the ‘Von Neumann Effect’. (Muir, 1988, cited by Nass, Moon, Fogg, Reeves & Dryer, 1995) Wetmore’s theory (1999) that humans apply anthropomorphism as a framework from within which to comprehend technology can certainly be observed in the technical language of computing. A computer has a ‘memory’, it ‘searches’ for things, it uses ‘languages’ and can suffer terribly from a ‘virus’. In addition to this there is now a substantial body of evidence to suggest that humans instinctively respond to their computers in a social manner. The man-machine environment elicits different responses than other media. When you read a newspaper, you are fully aware that you are reading the work of an author, you do not feel that the newspaper is speaking to you as an entity in it’s own right. Because of the nature of our interaction with technology there is a tendency to believe that we are dealing directly with another ‘person’ and not the author(s) of the program. When a computer goes wrong, it is generally easier for the user to say “it doesn’t like me” than it might be for them to work out what is really happening. Rickenberg & Reeves (2000) observe that people confer human personalities on the simplest of interface characters: “Rather than seizing on the differences between such characters and humans - a process which requires thought and deliberation – people slip into social conventions because important features of interactions with animated characters mimic real life….” Individuals often use inappropriate social rules in assessing machine behaviour (Nass, Steuer, Henriksen & Dryer 1994) In human-computer interaction, the term anthropomorphism is generally applied to describe this behaviour if the design of the computer is such that it convinces the user that it can actually carry out human interactions. However, if users are simply mindlessly applying their own social behaviours and overused human social categories to computers, whilst continuing to consciously reject the notion that the computer is capable of functioning as a human, then this is referred to as ethopoeia (Nass & Moon, 2000) In line with the theory of ethopoeia, Nass, Steuer, Henriksen & Dryer, (1994) report that individuals have often been observed applying social heuristics to evaluate the behaviour of machines that are inconsistent with their actual espoused beliefs about those machines. In the past, scholars have asserted that sustained social interaction with computers is based on user deficiency – either through ignorance or psychological or social dysfunction (Turkle, 1984, Winograd & Flores, 1987, Barley, 1988, Zuboff, 1988, Scherz, Goldberg & Fund, 1990 all cited by Nass, Steuer, Henriksen & Dryer, 1994) – or that individuals may have adopted an “intentional stance” toward their computer because it reflects the attitudes and intentions of the system designer. (Dennett, 1988, cited by Nass, Steuer, Henriksen & Dryer, 1994). However, recent studies have disputed these claims, by providing compelling evidence that individuals apply social interaction to their machines unconsciously. Many participants who are questioned after experiments emphatically deny the very behaviour that they have just exhibited (Reeves & Nass 1996) and these unconscious responses may be triggered by minimal social cues. Reeves and Nass speculate that humans may have evolved to assume that objects which exhibit certain humanlike traits are actually human. So, when modern humans are presented with interactive media they unconsciously respond to them in a social way, even though they consciously realise that their computer is not actually human. |
||||||
1.3 Computers Are Social Actors |
||||||
The “Computers Are Social Actors” paradigm was first proposed by Nass, Steuer & Tauber in 1994. Five experiments were carried out in an attempt to provide evidence that individual’s interactions with computers are fundamentally social. The first experiment investigated computers and politeness. During the de-briefing, many of the test subjects indicated their belief that the norms of politeness do not apply to computers. They were asked to carry out a series of tasks on a computer and to evaluate the computer’s performance – some subjects evaluated the performance on the same computer, some used pen and paper and some used a different computer. Those who used the same computer generally rated the computer’s performance higher than those who used pen and paper or a different computer. This suggests that users who used the same machine felt compelled to be more positive about the computer, out of ‘politeness’ because they considered that it would be ‘aware’ of their input. The second experiment established that users apply the notion of ‘self’ and ‘other’ to computers and the third experiment pinpointed the locus of this self/other attribution by using an electronic voice on different machines to establish that subjects respond to different voices as if they are distinct social actors and to the same voice as if it is the same social actor, regardless of which computer it is on. The final two experiments established that gender stereotypes from human-human interaction are also applied in human-computer interaction, and that the computer referring to itself as “I” is not necessary for the user to generate a social response. One aim of the study was to disprove previous theories that social interaction was caused by dysfunction, deficiency or a momentary aberration. The test subjects were experienced computer users, whom directly after declaring that the attribution of social rules to human-computer interaction is inappropriate, went on to unconsciously apply those rules in each of the tests. The authors conclude that these social responses are easy to generate, commonplace and incurable and that the human-computer relationship is fundamentally social. |
||||||
1.4 Computers with Personality |
||||||
Following the 1994 Computers Are Social Actors paper, Clifford Nass, a key advocate for applying social rules to human-computer interaction has co-written many papers continuing in the same vein of research. By applying the social psychological principles documented in human-human interaction to human-computer interaction, Nass and his colleagues have established that computer ‘personalities’ can be easily created, using a minimal set of cues (without the need for artificial intelligence) and that people will respond to these in the same way that they would respond to human personalities (Nass, Moon, Fogg, Reeves & Dryer, 1995). A further study (Nass, 1998), which addressed concerns that users may be abdicating too much responsibility to computers, found that similarity attraction extends to the user’s apportion of blame and responsibility, i.e. when users are working on a computer that they feel dissimilar to, they are more likely to blame that computer for failure, yet take the credit for success. Conversely if they are working on a computer which has a similar ‘personality’ to their own, they are more generous in their attributions, blaming themselves for failure and giving credit to the computer for any success. The author (ibid.) concludes that one possible solution to the abdication of responsibility would be to match different versions of a software program to the personalities of different users. They suggest that this would: “..not only make computers more enjoyable and easier to use, but would also create a class of more responsible users…." 1.4.2 The Complimentary Principle The personality types used were ‘introverted’ and ‘extroverted’ as opposed to the ‘dominant’ and ‘submissive’ types used in the previous two examples, this may have influenced the contrasting effects. The authors (ibid.) suggest that the theories of similarity-attraction (like attracts like) and the complimentary principle (opposites attract) are competing hypotheses in the field of social psychological research. The complimentary principle supports the idea that people will interact in complimentary ways and seek out others that elicit complimentary behaviour from them. This counteracts imbalance and tension by creating a balance of power within an interaction, where one person is more dominant than the other. Unlike the two previously mentioned examples, which used only text-based manipulations, the experiment (ibid.) involved an on-screen character, which displayed its personality through body language (non-verbal cues) as well as through text (verbal cues). The authors suggest that this may have influenced the results as the on-screen character may have increased the user’s sense of social presence and a direct interaction. These experiments show that although humans have no problem in attributing human characteristics and ‘personalities’ to computers, the choice of personality type may be crucial to the success of the application. It is highly unlikely that there is one personality type that will be universally liked. In the book, The Media Equation (Reeves & Nass, 1996: 98) it is suggested that if a program may have only one personality, that the designer should pick a personality type which ‘suits’ the application – for example if the computer is an active helper, then the personality should be dominant and friendly. Ideally, if the computer is able to have more than one personality then a selection of ‘personality types’ should be available for the user to choose, prior to commencement. In addition to the choice of dominant/submissive and extrovert/introvert personality representations, several studies have been completed which investigate the effects of other human characteristics upon user interaction. A study exploring the use of humour in task-oriented interactions (Morkes, Kernal & Nass 1998) found that participants who had received jokes: “rated the ‘person’ or computer they had worked with as more likeable and competent, reported greater cooperation, joked back…….and smiled and laughed more”. The study noted that computer programmers tend to dismiss humour because of a fear of ‘process losses’ (i.e. distractions from the task at hand). They contend that the computer is a tool and as such should be designed to minimise diversions from the goal of completing a task. This does not take into account the fact that some users actually welcome an occasional distraction as some computer-based tasks can be exceedingly mundane. The authors observe that people commonly use humour when they interact at work and that they frequently respond to computers in the same way that they might respond to their colleagues. They suggest that humour is powerful and can have many positive effects. As with the previous studies, artificial intelligence was not deemed necessary to produce the desired outcome and text-based ‘canned humour’ proved to be sufficient. Similarly, an experiment which examined the ‘effects of computers that flatter’ (Fogg & Nass, 1997), revealed that “praise has multiple beneficial effects on user perceptions and that flattery from computers generates the same effects as sincere praise…..Humans are susceptible to flattery from computers in the same way that they are susceptible to flattery from other humans.” The authors (ibid.) suggest that this could be seen as an opportunity to enhance the user experience by using flattery to lift the mood of users and improve their perceptions of the machine. Current applications tend to be heavily geared towards critical messages, if sincere praise or flattery were added to, for example, a tutorial application, it is proposed that it may well increase user enjoyment, task persistence and self-efficacy, which are vital conditions in a learning environment. In 2004, a study by Jeng-Yi Tzeng at the National Tsing Hua University in Taiwan, looked at the effects of computers which apologise. Inspired by the Chinese saying, “no-one would blame a polite person” Tzeng created two versions of a computer-based guessing game, one which apologised and one which did not. The games were then tested on 269 high-school students who were given the impression that the study was intended to determine the playability of each game. The study found that the apologetic messages did not tangibly improve the subject’s actual achievement, or divert them from feeling bad about their performance, but they did make the game more aesthetically desirable. The messages placed the computer in a more favourable light, it was viewed as being less mechanical, more enjoyable to play with and more sensitive to the user’s emotions. “It is exactly what I would have expected…..as computers have got more powerful, people have come to expect them to behave more like collaborators and less like tools or appliances…” Despite the commercial failure of anthropomorphic products such as the ‘Office Assistant (aka ‘Clippy’ the Paperclip) and ‘Microsoft Bob’ (see Chapter Two), Microsoft have remained committed to the idea of ‘the social interface’ and using interface agents to make computers “100 times easier to use than today’s VCR is” – Bill Gates (1995) Yet not everyone is so keen to embrace our ‘instinctive’ and ‘unconscious’ tendencies towards anthropomorphism. |
||||||
1.5 The Opposition |
||||||
In his 1995 paper, ‘Agents of Alienation’, Jaron Lanier declares that the idea of an intelligent agent (such as an anthropomorphic interface) is “both wrong and evil”. He describes the Turing test as the “creation myth of artificial intelligence”. Asserting that it has been taken out of context, Lanier explains: “Alan Turing…happened to be gay and was imprisoned by the British government and subjected to quack treatments for homosexuality….[he] was trying to escape the pain of his circumstances by fantasising an abstract intelligence, free of the dreadful miseries of the flesh….it was under these conditions, shortly before his suicide, that he created the mythical basis for smart machines..” Another fierce opponent of anthropomorphism in interface design is Ben Shneiderman, who describes anthropomorphism in computing as “deceptive, counterproductive and morally offensive to me” (Brennan, Laurel & Shneiderman, 1992). He suggests that every technology “passes through an immature phase in which human and animal models are used as metaphors for design”. In an article for Forbes magazine (1998) he goes on to claim that artificial intelligence has been a counterproductive and misdirected effort for 30 years and refers to it as the ‘obstacle of animism.’ In Designing the User Interface (Shneiderman & Plaisant, 2004: 484), Shneiderman sets out his three main concerns regarding the use of anthropomorphism in interface design: Firstly, he believes that designing an interface that suggests that a computer is intelligent, autonomous or capable of understanding, can confuse or mislead and is ultimately deceiving the user. Once the user realises this deception they may resent or dislike the computer. Watt (1998, cited by Swartz, 2003) refers to the gap between expected behaviour and actual behaviour as “anthropomorphic dissonance” – and claims that the bigger the gap, the greater the dissatisfaction with the interface. Shneiderman’s second concern is that the boundary between people and computers should be clarified, not blurred. He states that “users and designers must accept responsibility for misuse of computers rather than blaming the machine for errors.” (Shneiderman & Plaisant, 2004: 485). In a personal communication (2005) he reiterates this point, stating: “People much prefer to be in control of their computers (and are more productive in these situations) than to have a social relationship with their computers. In short, human-human interaction is rarely a good source of inspiration for human-computer interaction. People are NOT computers; computers are NOT people.” In Tzeng’s study of computers that apologise (2004) He responds to Shneiderman’s concerns about responsibility with what he claims is “soothing evidence” that people are less likely to blame the computer if it is polite (ie anthropomorphic). He asserts that this is because interacting with computers triggers a social schema at a sub-conscious level, which causes the user to behave in a civilised way and makes them reluctant to ‘hurt the computer’s feelings.’ The third concern expressed by Shneiderman is that although some people might prefer the anthropomorphic interface, it may cause anxiety for others. In a personal communication (2005) he stated that he had no objection to a company building an anthropomorphic interface to give users a choice, but he thought that they would be more commercially successful if they devoted their energy to advanced designs that improved user productivity and reduced frustration. In an earlier edition of ‘Designing the User Interface’ (Shneiderman, 1992: 383), Shneiderman refers to a study by Resnik and Lammers (1986) which indicated that an interface which used anthropomorphic messages (such as ‘I don’t understand these numbers’) was more likely to make users confused and nervous than one which used constructive text (such as ‘use letters only’). Shneiderman is an enthusiastic advocate of ‘Direct Manipulation’, a term that he originated in 1983, which is used to describe a method for controlling a computer by directly manipulating the interface elements. Shneiderman’s grave objections to anthropomorphism have led some people to view the issue as a debate between the two forms of interface. Brenda Laurel (Laurel, 1992) maintains that Shneiderman supports direct manipulation over anthropomorphism for several reasons: “Ben is comfortable with tool-like interfaces, because he works primarily with tool-like applications. But one can no more confine computers to applications like spreadsheets and word processors than one can could confine the printing press to Latin versions of the bible…” Brennan and Ohaeri (1994) conclude in their study on message style that it is counterproductive to pit direct manipulation interfaces against anthropomorphic ones. Each concept has its own distinct advantages and they assert that the focus of understanding should be on how to get the most out of both. In an earlier paper (Brennan, Laurel & Shneiderman, 1992) Brennan states that just because certain aspects of anthropomorphism, such as imitating a human, have been proven to be misleading, that is not a good reason to completely discount it as a viable way of enhancing interface use. Brennan (ibid.) claims that those who debate the value or evils of anthropomorphic interfaces are missing the point. So, the evidence presented here certainly suggests that people treat their computers as social actors (even when they have denied doing so). But Shneiderman’s arguments against embracing these instinctive actions are very compelling. When questioned on his views on the CASA theory he responded: “As for CASA – my dear colleague Cliff Nass is a clever guy, but I think the theory is incomplete. His studies do show that users will respond socially, but he does not adequately consider whether they will be happier or more productive with non-anthropomorphic interfaces. I might respond with the counter-theory of SACA, Sugar as a Culinary Actor. People may prefer and be willing to eat breakfast cereals with more sugar, but it does NOT mean that sugar is good for them, or the best way to make healthy cereals” (2005) Certainly his arguments are supported by the historical failures of anthropomorphic interfaces such as the Microsoft Office Assistant (aka Clippy) and the empirical failures to demonstrate enhanced productivity or usability (rather than just user preference) in controlled studies. The next chapter investigates previous attempts at producing a humanlike interface, (such as interface agents) and examines the factors which may have affected their commercial success. Is it simply a case of the wrong personality – or are they truly “wrong and evil” (Lanier, 1995)? |
||||||
© Alison Flind 2006 |